U.S. patent application number 13/881170 was filed with the patent office on 2013-11-07 for rapid method for targeted cell (line) selection.
This patent application is currently assigned to LONZA BIOLOGICS PLC. The applicant listed for this patent is Dietmar Lang, Elaine B. Martin, Gary A. Montague, Christopher J. O'Malley, Jane F. Povey, Andrew J. Racher, Tracy S. Root, Christopher M. Smales, Carol M. Trim. Invention is credited to Dietmar Lang, Elaine B. Martin, Gary A. Montague, Christopher J. O'Malley, Jane F. Povey, Andrew J. Racher, Tracy S. Root, Christopher M. Smales, Carol M. Trim.
Application Number | 20130295596 13/881170 |
Document ID | / |
Family ID | 43639936 |
Filed Date | 2013-11-07 |
United States Patent
Application |
20130295596 |
Kind Code |
A1 |
Lang; Dietmar ; et
al. |
November 7, 2013 |
RAPID METHOD FOR TARGETED CELL (LINE) SELECTION
Abstract
The present invention relates to a process for the prediction of
cell culture performance data of sample cells, a process for the
isolation of said cells and a device for the prediction of cell
culture performance data of sample cells.
Inventors: |
Lang; Dietmar; (Liverpool,
GB) ; Martin; Elaine B.; (South Gosforth, GB)
; Montague; Gary A.; (Newcastle, GB) ; O'Malley;
Christopher J.; (Newcastle, GB) ; Root; Tracy S.;
(Slough, GB) ; Trim; Carol M.; (Ramsgate, GB)
; Povey; Jane F.; (Deal, GB) ; Smales; Christopher
M.; (Chartham, GB) ; Racher; Andrew J.;
(Brightwalton, GB) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Lang; Dietmar
Martin; Elaine B.
Montague; Gary A.
O'Malley; Christopher J.
Root; Tracy S.
Trim; Carol M.
Povey; Jane F.
Smales; Christopher M.
Racher; Andrew J. |
Liverpool
South Gosforth
Newcastle
Newcastle
Slough
Ramsgate
Deal
Chartham
Brightwalton |
|
GB
GB
GB
GB
GB
GB
GB
GB
GB |
|
|
Assignee: |
LONZA BIOLOGICS PLC
Slough, Berkshire
GB
|
Family ID: |
43639936 |
Appl. No.: |
13/881170 |
Filed: |
October 27, 2011 |
PCT Filed: |
October 27, 2011 |
PCT NO: |
PCT/EP11/05407 |
371 Date: |
July 9, 2013 |
Current U.S.
Class: |
435/23 ;
435/287.1; 435/358 |
Current CPC
Class: |
H01J 49/164 20130101;
G01N 33/5005 20130101; G01N 33/6848 20130101; H01J 49/40 20130101;
H01J 49/0031 20130101; G01N 27/62 20130101 |
Class at
Publication: |
435/23 ; 435/358;
435/287.1 |
International
Class: |
G01N 27/62 20060101
G01N027/62 |
Foreign Application Data
Date |
Code |
Application Number |
Oct 27, 2010 |
EP |
10014005.2 |
Claims
1-12. (canceled)
13. A process for the prediction of cell culture performance data
of at least one sample cell, the process comprising: (a) providing
a sample of the at least one sample cell, cell culture performance
data from a standard cell and raw standard MS (mass spectrometric)
data from the standard cell, (b) subjecting the sample of the at
least one sample cell to a MS analysis to obtain raw sample MS data
thereof, (c) subjecting the raw standard and the raw sample MS data
to at least one first MS signal processing method to obtain
pre-treated standard and sample MS profiles and (d) subjecting the
cell culture performance data from the standard cell of (a) and the
pre-treated standard and sample MS profiles obtained in (c) to a
second MS signal processing method including a PLS-DA (partial
least squares discriminant analysis) based comparative evaluation
so as to predict the cell culture performance data of the at least
one sample cell, wherein the cell culture performance data are cell
specific productivity, integral viable cell count or cell product
concentration data.
14. The process according to claim 13, wherein the cell is selected
from a group consisting of human cell lines, animal cell lines,
plant cell lines, cells from fungi, cells from bacteria, cells from
yeast and stem cells.
15. The process according to claim 13, wherein the cell is a CHO
cell line or a CHO-K1 cell line.
16. The process according to claim 13, wherein the MS analysis in
(b) is MALDI-TOF.
17. The process according to claim 13, wherein the sample of the
sample cells subjected to (b) comprises from 0.015.times.106 to
0.0625.times.106 cells.
18. The process according to claim 13, wherein the raw sample MS
data obtained in (b) and the raw standard MS data provided in (a)
are signal processed by an operation selected from the group
consisting of baseline correction, normalisation, alignment,
filtering and cropping.
19. The process according to claim 13, wherein the pre-treated
standard and sample MS profiles obtained in (c) are optically
analysed.
20. The process according to claim 13, wherein the sample cell with
the predicted cell culture performance data evaluated in (d) is
cultivated in a cell culture so as to verify its cell culture
performance data.
21. The process according to claim 20, wherein the raw sample MS
profiles obtained in (b) and the verified cell culture performance
data of the sample cell are used in (a) as standard MS data and
cell culture performance data from a standard cell.
22. A method of isolating a cell with desired cell culture
performance data comprising: predicting cell culture performance
data of at least one sample cell with the process of claim 13; and
isolating a desired cell having a desired cell culture performance
data.
23. A cell isolated by the process according to claim 22, wherein
the cell has a protein productivity of at least 10 g/l/h.
24. A device that is adapted to provide a prediction of cell
culture performance data when supplied with a sample of at least
one sample cell, the device comprising the: (a) means adapted for
subjecting a sample of the at least one sample cell to a MS (mass
spectrometric) analysis to obtain a raw sample MS data thereof; (b)
means adapted for subjecting a raw standard and the raw sample MS
data to at least one first MS signal processing method to obtain
pre-treated standard and sample MS profiles; and (c) means adapted
for subjecting cell culture performance data from a standard cell
and the pre-treated sample and standard MS profiles to a second MS
signal processing method including a PLS-DA (partial least square
discriminant analysis) based comparative evaluation so as to
predict the cell culture performance data of the sample cell,
wherein the cell culture performance data are cell specific
productivity, integral viable cell count or cell product
concentration data.
Description
[0001] The present invention relates to a process for the
prediction of cell culture performance data of sample cells, a
process for the isolation of said cells and a device for the
prediction of cell culture performance data of sample cells.
[0002] According to the prior art methods for isolating recombinant
mammalian cell lines with desired manufacturing properties are
inefficient both in terms of resources and the ability to isolate
specific combinations of desired characteristics.
[0003] For instance, Adrichem et al. (Anal. Chem., 1998, 70,
923-930) disclose investigations of protein patterns and mammalian
cells in culture supernatants by matrix-assisted laser
desorption/ionisation mass spectrometry (MALDI mass spectrometry).
Said MALDI mass spectrometry can be used for protein profiling by
monitoring proteins which have either been excreted into the media
or located in cell lysates. The detectable mass range by MALDI mass
spectrometry is from 16000 to several hundred thousands of Daltons,
preferably from several hundred to several thousand. The results
obtained thereby are complementary with standard SDS-PAGE
electrophoresis. Therefore, these methods can be used for e.g.
monitoring a large scale cultivation of hybridoma cells expressing
an antibody of the IgG type.
[0004] Zhang et al. (J. Am. Soc. Mass Spectrom., 2006, 17, 490-499)
disclose the identification of mammalian cell lines using MALDI-TOF
and LC-ESI-MS/MS mass spectrometry. It is stated that MALDI-TOF
mass spectrometry is effective for peptide profiling directly from
single cell. LC-ESI-MS/MS analysis can provide useful sequence
information after a tryptic digestion of sample cells. It was found
to yield unique and reproducible MALDI-MS patterns, which can be
used as a fingerprint to identify and distinguish the different
cell lines measured. However, the obtained MS spectra were compared
visually on basis of the clear visible MS peaks to distinguish the
three different mammalian cell types. Alternatively, Zhang et al.
demonstrate that a different technique--a combination of liquid
chromatography followed by electrospray ionisation and tandem mass
spectroscopy (LC-ESI MS/MS) for which it is necessary to digest the
samples--is also useful to shed light on proteome profile
differences.
[0005] Feng et al. (Rapid Commun. Mass Spectrom., 2010, 24,
1226-1230) disclose a rapid characterisation of high/low producer
CHO cells using matrix-assisted laser desorption/ionisation
time-of-light (MALDI-TOF). The process disclosed therein is able to
distinguish between high and low producer cells when produced in
the same culture at the same scale by applying two statistical
methods, namely principle component analysis (PCA) and linear
partial least squares (PLS), to analyse the MALDI-TOF spectra.
Especially, the method according to Feng et al. allows
distinguishing between productivity data from different cell lines
producing a recombinant protein IFN-gamma at the same scale, i.e.
grown at low scale. According to Feng et al. this approach could
possibly be used to predict cell productivity. The linear PLS used
derives its usefulness from the ability to analyse data with many
variables which is relevant to find cell lines subfamilies at a
given scale.
[0006] As mentioned above the methods according to the prior art,
especially to Feng et al., teach merely that MALDI-TOF data being
analysed by statistical programs, namely PLS and PCA, can be used
for differentiating between low and high producer cells at a given
scale of culturing.
[0007] However, the prior art fails to teach a method which
predicts the cell characteristics of an unknown cell at a later
stage of up-scaling, in particular in large volume bioreactors,
while these cells are still cultivated in a medium with a low
volume. Especially, the known statistical programmes of PCA and PLS
produce data being insufficient to be used as a basis for an
accurate and reliable prediction of cell characteristics at a later
state of up-scaling.
[0008] A cell line at the early stage expressing a specific balance
of proteins can exhibit for instance a high productivity at this
stage, but after up-scaling to bioreactor stage this productivity
can be deteriorated due to different factors, which are inter alia
shear forces, volume effects, different fermenter type or format,
cultivation parameters, for example pH and gas-controlled cell
density differences.
[0009] Therefore, there is a need for a method having the ability
to rapidly screen large numbers of cell lines early in cell line
development and cultured in small volumes only, which is able to
predict the growth and specific production characteristics, for
instance increased volumetric productivities, of said cell line at
a given and desired culture scale, in particular in a bioreactor
scale, that means in large volume scale.
[0010] Thus, the technical problem underlying the present invention
is to provide a method to overcome the above identified problems,
in particular to provide a method for the prediction of the cell
culture, in particular bioreactor, performance of a cell with
unknown cell characteristics, in particular already at an early
state of up scaling, being additionally time-saving and
cost-reducing, but having a high prediction accuracy for said
performance in a large volume scale cultivation and/or under high
productivity conditions, in particular in a bioreactor scale.
[0011] This problem is solved by the independent claims of the
present invention.
[0012] Thus, the present invention provides a process, preferably
an in vitro process, for the prediction of cell culture performance
data of at least one sample cell comprising the steps: [0013] a)
providing a probe of the at least one sample cell, preferably
cultivated in a low volume of a medium, cell culture performance
data from a standard cell and raw standard MS (mass spectrometry)
data from a standard cell, [0014] b) subjecting the probe of the at
least one sample cell to a MS analysis to obtain a raw sample MS
data thereof, [0015] c) subjecting the raw standard and the raw
sample MS data to at least one first MS signal processing method to
obtain pre-treated standard and sample MS profiles and [0016] d)
subjecting the cell culture performance data from the standard cell
from step a) and the pre-treated sample and standard MS profiles
obtained in step c) to a second MS signal processing method
including, preferably to a data analysis comprising of, a PLS-DA
(partially least squares-discriminant analysis) based comparative
evaluation so as to predict the cell culture performance data of
the at least one sample cell.
[0017] The present invention therefore provides an advantageous
process for the accurate and reliable prediction of cell culture,
preferably bioreactor, performance data of at least one sample cell
with unknown cell culture, preferably bioreactor, performance data,
which allows the time saving and cost-reducing prediction of
particular cell culture, preferably bioreactor, performance data,
such as the productivity of cells. The present invention not only
provides such an advantageous process, but also provides a cell
prepared, and in particular isolated, by said method, wherein said
cell is particularly characterised by desired cell culture,
preferably bioreactor, performance data, such as a high
productivity. Furthermore, the present invention provides a device
for the prediction of cell culture performance data capable of
conducting the process of the present invention.
[0018] In contrast to the prior art which usually upscales from a
96 well plate first to a 24 well plate, then to a shake flask and
finally to bioreactor scale, the present invention obviates the
cultivation steps in a 24 well plate and in a shake flask. Directly
after the prediction of the cell culture, preferably bioreactor,
performance data of at least one sample cell cultivated in low
volume, for instance in a 96 well plate, the cultivation in a
bioreactor can be performed.
[0019] According to the present invention, the PLS-DA allows the
separation of very specific classes of observations on the basis of
one variable, so that with the use of PLS-DA the problem of
productivity classification for cell lines is overcome and their
cell culture, preferably bioreactor, performance at different scale
can be predicted.
[0020] In the context of the present invention the cell culture
performance data are preferably bioreactor performance data.
[0021] In the context of the present invention the term "a probe of
the at least one sample cell" has the same meaning as the term "a
sample of the at least one sample cell".
[0022] In the context of the present invention the term "bioreactor
performance data" is understand to mean data on the behaviour and
the characteristics of a cell, when said cell is cultivated or
reproduced under a large volume condition and/or high productivity
conditions, in particular in a bioreactor. The bioreactor
performance data are preferably data on the individual and specific
cell productivity for a specific cell product, e.g. protein, in
particular antibody, antibody fragment or fused antibody,
antibiotics, cultivation needs, growth or lifetime of said cell. In
a preferred embodiment, the cell product is a protein, in
particular an antibody, peptide, proteoglycan, glycoprotein,
carbohydrate, lipid, antibiotic or hormone.
[0023] The term "standard cell" means a cell with known cell
culture performance data, in particular known characteristics and
behaviour at a given scale, preferably at a large volume scale
and/or under high productivity conditions, in particular at a
bioreactor scale. These known characteristics were measured and
analysed by methods according to the state of the art. For
instance, the cell productivity can be determined by ELISA
(enzyme-linked immuno sorbent assay).
[0024] In the context of the present invention the term "sample
cell" relates to a cell having at least one unknown cell culture
performance data, in particular cell characteristic. Preferably, at
least one characteristic of said sample cell is known.
[0025] In the context of the present invention the term "low volume
of a medium" means that the medium preferably has a volume of
preferably 1 .mu.l to 100 l, preferably 1 .mu.l to 90 l, preferably
1 .mu.l to 80 l, preferably 1 .mu.l to 70 l, preferably 1 .mu.l to
60 l, preferably 1 .mu.l to 50 l, preferably 1 .mu.l to 40 l,
preferably 1 .mu.l to 30 l, preferably 1 .mu.l to 20 l, preferably
1 .mu.l to 10 l, preferably 1 .mu.l to 5 l, preferably 1 .mu.l to 4
l, preferably 1 .mu.l to 3 l, preferably 1 .mu.l to 2 l, preferably
1 .mu.l to 1 l, preferably 1 .mu.l to 0.5 l, preferably 1 .mu.l to
0.4 l, preferably 1 .mu.l to 0.3 l, preferably 1 .mu.l to 0.2 l,
preferably 1 .mu.l to 0.11, preferably 1 .mu.l to 90 ml, preferably
1 .mu.l to 80 ml, preferably 1 .mu.l to 70 ml, preferably 1 .mu.l
to 60 ml, preferably 1 .mu.l to 50 ml, preferably 1 .mu.l to 40 ml,
preferably 1 .mu.l to 30 ml, preferably 1 .mu.l to 20 ml,
preferably 1 .mu.l to 10 ml, preferably 1 .mu.l to 5 ml, preferably
1 .mu.l to 4 ml, preferably 1 .mu.l to 3 ml, preferably 1 .mu.l to
2 ml, preferably 1 .mu.l to 1 ml, preferably 1 to 999 .mu.l,
preferably 1 .mu.l to 0.5 ml, preferably 1 .mu.l to 0.4 ml,
preferably 1 .mu.l to 0.3 ml, preferably 1 .mu.l to 0.2 ml,
preferably 1 .mu.l to 0.1 ml, preferably 1 .mu.l to 50 .mu.l,
preferably 1 .mu.l to 40 .mu.l, preferably 1 .mu.l to 30 .mu.l,
preferably 1 .mu.l to 20 .mu.l, preferably 1 .mu.l to 10 .mu.l,
preferably from 20 to 90 .mu.l, preferably 20 to 80 .mu.l,
preferably from 20 to 70 .mu.l, preferably from 20 to 60 .mu.l,
preferably from 20 to 50 .mu.l, preferably from 20 to 40 .mu.l,
preferably from 20 to 30 .mu.l, preferably from 30 to 70 .mu.l,
preferably from 30 to 60 .mu.l, preferably from 30 to 50 .mu.l,
preferably from 30 to 40 .mu.l, preferably from 10 to 100 .mu.l,
preferably from 10 to 90 .mu.l, preferably from 10 to 80 .mu.l,
preferably from 10 to 70 .mu.l, preferably from 10 to 60 .mu.l,
preferably from 10 to 50 .mu.l, preferably from 10 to 40 .mu.l,
preferably from 10 to 30 .mu.l, preferably from 10 to 20 .mu.l,
preferably from 40 to 60 .mu.l, preferably from 40 to 50 .mu.l.
[0026] In the context of the present invention the term "early
stage of up-scaling" is understood as a point of time where the
cells are cultivated in a low volume of a medium as defined
above.
[0027] In the context of the present invention the term
"bioreactor" means a container capable of containing cells for the
production of at least one desired cell product, which preferably
enables a high productivity in terms of the production speed and/or
amount of said desired cell product. In a particularly preferred
embodiment the bioreactor is a device or system supporting a
biologically active environment. In a particularly preferred
embodiment a bioreactor of the present invention is a container
suitable for industrial and commercial production of said cell
product of interest. In a particularly preferred embodiment such a
container is able to create cell culture conditions suitable for
producing the cell product of interest with a high productivity. In
a particularly preferred embodiment the bioreactor contains a
volume of medium of at least 10 l, at least 20 l, at least 50 l, at
least 100 l, at least 200 l, at least 300 l, at least 400 l, at
least 500 l, at least 600 l, at least 700 l, at least 800 l, at
least 900 l, at least 1000 l, in particular at least 2000 l, at
least 3000 l or at least 4000 l, which is preferably termed to be a
"large volume" or "large volume of a medium" as used herein.
[0028] In the context of the present invention the expression
"and/or" used in between two elements is meant to designate that
both elements linked by said term are referred to in a cumulative
or alternative manner. Thus, the expression "A and/or B"
encompasses the meanings "A or B" and "A and B", that means "any
one of A, B or both".
[0029] Thus, the present invention provides a method which enables
a person skilled in the art to predict the characteristics of at
least one sample cell, especially at least one cell line, by
providing in a first step a) a probe of at least one sample cell
cultivated in a medium having a low volume and low cell
concentration therein, preferably from 10.sup.4 to 10.sup.8,
preferably from 10.sup.4 to 10.sup.8 viable cells/ml culture
medium, the cell culture performance data of a standard cell and
the raw standard MS data thereof. Subsequently, in a second step
the at least one sample cell is analysed by a mass spectrometry
method. In a third step both the raw MS data of the at least one
sample cell and the standard cell are treated by a MS signal
processing method so as to obtain pre-treated MS profiles of the at
least one sample cell and standard cell. In a fourth step the cell
culture performance data from the standard cell and the pre-treated
sample and standard MS profiles are subjected to a statistical
method, namely PLS-DA, preferably in combination with PCA. This
statistical program is employed to compare and evaluate the
pre-treated MS profiles of the sample cell to the pre-treated MS
profiles of a standard cell.
[0030] In a preferred embodiment of the present invention in step
a) a sample of the at least one sample cell is provided in order to
be subjected to a MS analysis to obtain a raw sample MS data
thereof in step b).
[0031] The term "a probe of the at least one sample cell" and the
term "a sample of the at least one sample cell" means a defined
number of sample cells, namely at least one sample cell.
[0032] Preferably, the sample is provided in step a) in a low
volume of a medium and subjected to a MS analysis to obtain a raw
sample MS data thereof.
[0033] In a preferred embodiment of the present invention, the
concentration of the cells, preferably in the sample, is from
10.sup.4 to 10.sup.8, from 10.sup.5 to 10.sup.7, in particular from
10.sup.6 to 10.sup.7 viable cells/ml culture medium after finishing
the cultivation.
[0034] The present invention enables the prediction of the cell
culture, preferably bioreactor, performance data, e.g. the
productivity, of the at least one sample cell being cultivated at
an early stage of up-scaling. The method according to the present
invention allows the rapid development of for instance highly
productive cell lines and thereby reduces the costs of therapeutic
protein manufacturing and speeds up the development of
pharmaceuticals.
[0035] The present process provides a novel screening tool for
identifying, preferably early in the product development cycle,
i.e. in a phase where a culture medium with a volume of preferably
1 to 999 .mu.l is used, cells, preferably cell lines, that have the
desired properties, for instance a high volumetric productivity, in
particular in large-scale bioreactors, in particular in bioreactors
containing culture medium in a volume above 10 litre. Moreover, the
method according to the present invention improves the probability
of finding a cell, especially cell line, with specific cell
characteristics, for instance comprising a high product
productivity, preferably at least 1 g/l, preferably at least 5 g/l
and most preferably 10 g/l, early in development. In addition to
identifying valuable cells, preferably cell lines, the method can
also be used preferably for isolating new host cells, preferably
cell lines, with improved properties for instance for therapeutic
protein manufacturing, especially monoclonal antibody
manufacturing. In particular, by the combination of the MS analysis
with the statistical method of PLS-DA, preferably by the
combination of PLS-DA with PCA (principle component analysis), the
present process allows the identification of patterns in different
product productivity levels as to predict the productivities at
different scale therefrom in a fast and, if applicable, an
automatic way.
[0036] In a preferred embodiment of the present invention the cell
specific productivity, preferably high product productivity of the
cells, is at least 0.1 g/l/h, preferably at least 1 g/l/h,
preferably at least 5 g/l/h, preferably at least 10 g/l/h and most
preferably 10 g/l/h.
[0037] The high probability of finding high producing cells,
preferably cell lines, has the potential to reduce the number of
cells, preferably cell lines, that needed to be screened before one
suitable for manufacturing is identified. By reducing the number of
screened cells, preferably cell lines, the materials required are
reduced. Consequently, fewer resources will be required during the
cell, preferably cell line, construction with concomitant
generation of less waste materials. Higher producing cells,
preferably cell lines, will reduce the number of bioreactor
cultures required to supply the market requirements for the
product. The present process is able to reduce the raw material
requirements, in particular costs, of the production process,
especially water that is used both as a raw material and in
cleaning and sanitisation of the equipment.
[0038] In a preferred embodiment of the present invention the cell
culture performance data, preferably the bioreactor performance
data, reflects the performance of the cells, preferably of the
sample cells and/or the standard cells, in a large volume.
[0039] In a preferred embodiment of the present invention the cell
culture, preferably bioreactor, performance data is cell specific
productivity, integral viable cell count or cell product
concentration, in particular are cell specific productivity,
integral viable cell count or cell product concentration data.
[0040] In a particularly preferred embodiment the cell culture,
preferably bioreactor, performance data is the cell specific
productivity (qP).
[0041] In a particularly preferred embodiment the cell culture,
preferably bioreactor, performance data is the integral viable cell
count data (IVC).
[0042] In a particularly preferred embodiment the cell culture,
preferably bioreactor, performance data is the cell product
concentration data.
[0043] In a preferred embodiment of the present invention the cell
product concentration data is the titre data of the cell
product.
[0044] The term "titre" means the concentration of a medium,
preferably of a cell culture medium, preferably in a bioreactor, as
determined by titration.
[0045] In a preferred embodiment of the present invention data on
the production stability of cells is not understood as cell culture
performance data.
[0046] In a preferred embodiment of the present invention the
sample cell and/or standard cell is selected from the group
consisting of human cell lines, animal cell lines, plant cell
lines, antibodies, cells from fungi, cells from bacteria, cells
from yeast and stem cells.
[0047] In a preferred embodiment of the present invention both the
sample cell and the standard cell is selected from the same cell
type, especially cell line or strain.
[0048] In a preferred embodiment of the present invention the
sample cell is a CHO cell line, preferably a CHO-K1 cell line,
preferably a modified CHO-K1 cell line. In a preferred embodiment
of the present invention the standard cell is also a CHO cell line,
preferably a CHO-K1 cell line (ATCC Number: CCL-61.TM.), preferably
a modified CHO-K1 cell line.
[0049] A CHO-K1 cell line is a subclone of the parental CHO cell
line, which was derived from the ovary of an adult Chinese hamster.
CHO-K1 cells require proline due to the absence of the gene for
proline synthesis, with the block in the biosynthetic chain
occurring in the step converting glutamic acid to glutamine
.gamma.-semialdehyde.
[0050] In a preferred embodiment of the present invention at least
one part, preferably all of the standard cells express different
cell products, preferably proteins, preferably antibodies compared
to the at least one sample cell.
[0051] In a preferred embodiment of the present invention the cell
line construction of at least one part, preferably all of the
standard cells is different to that of the at least one sample
cell.
[0052] In a preferred embodiment of the present invention the MS
analysis used in step b) is selected from the group consisting of
MALDI-TOF, LC-ESI-MS (liquid chromatography electrospray ionisation
mass spectrometry) and LC-ESI-MS/MS (liquid chromatography coupled
to tandem mass spectrometry with electrospray ionization).
[0053] In a preferred embodiment of the present invention the MS
analysis used in step b) is MALDI-TOF. The sample cells subjected
to MALDI-TOF need not to be digested by e.g. trypsin, but can be
embedded into the matrix in intact form by a very simple
preparation of the cells.
[0054] In a preferred embodiment of the present invention the MS
analysis used in step b) is LC-ESI-MS. LC-ESI-MS analysis provides
a particular great discrimination between cell lines in terms of
productivity, growth and other desirable characteristics. The
LC-ESI-MS analysis provides an extra dimension of information.
[0055] In a preferred embodiment of the present invention the raw
standard MS data is obtained by MS analysis selected from the group
consisting of MALDI-TOF, LC-ESI-MS and LC-ESI-MS/MS, preferably
MALDI-TOF.
[0056] In a preferred embodiment of the present invention, the
ionisation used for MALDI-TOF MS or LC-ESI-MS is carried out in a
negative or positive reflection mode or in a positive or negative
linear mode being optimal according to instrument-specific
parameters, for example being device-dependent, with or without
mass suppression and pulsed ion extraction.
[0057] In a preferred embodiment of the present invention, the
following settings of the MALDI-TOF mass spectrometer instrument
are used for the method according to the present invention:
Polarity: +ve
Suppress at: 1000 Da (Daltons)
Range: 1000 to 50000 Da
[0058] In a preferred embodiment of the present invention the mass
suppression during the MS analysis of MALDI-TOF and LC-ESI-MS is
below 500 Da, preferably below 1000 Da and most preferred below
1500 Da.
[0059] In a preferred embodiment of the present invention the mass
suppression during the MS analysis of MALDI-TOF and LC-ESI-MS is
above 500 Da, preferably above 1000 Da and most preferred above
1500 Da.
[0060] In a preferred embodiment of the present invention the
detected range during the MS analysis of MALDI-TOF and LC-ESI-MS is
200 to 100000 Da, preferably 200 to 60000 Da, preferably 500 to
50000 Da, preferably 1000 to 100000 Da, 1000 to 18000 Da,
preferably 500 to 10000 Da and most preferably 200 to 8000 Da.
[0061] In a preferred embodiment of the present invention, the at
least one sample cell is washed, preferably with a buffer solution,
before being subjected to the MS analysis, preferably to
MALDI-TOF.
[0062] In a preferred embodiment of the present invention the at
least one sample cell is washed either with phosphate buffered
saline (PBS) alone or followed by an aqueous sucrose solution wash
step, in particular with 0.2 to 0.7 M, preferably 0.3 to 0.5 M,
preferably 0.35 M sucrose before being subjected to the MS
analysis, preferably to MALDI-TOF.
[0063] In a preferred embodiment of the present invention, the
matrix used for the MALDI-TOF analysis is sinapinic acid (SA).
According to the present invention the use of sinapinic acid (SA)
as a matrix for the MALDI-TOF MS analysis provides advantageous
spectra with a particular wide range of peaks, preferably up to 70
kDa, and being particularly well resolved. In a further embodiment
of the present invention 2,5-dihydroxybenzoic acid (DHB) can also
be used as a matrix.
[0064] In a preferred embodiment of the present invention the probe
of the at least one sample cell subjected in step b) comprises not
more than 1.times.10.sup.6 cells, preferably 0.015.times.10.sup.6
to 0.0625.times.10.sup.6 cells, in particular 0.03.times.10.sup.6
cells.
[0065] In a preferred embodiment of the present invention the MS
profiles are taken after 1 to 5 hours, preferably 1 to 4 hours, in
particular 1 to 3 hours of acclimatisation at low temperature, for
instance at 0 to 10.degree. C., preferably at 2 to 8.degree. C., in
particular at 4.degree. C., resulting in the best reproducibly and
a lower signal to noise-containing mass spectra.
[0066] In a preferred embodiment of the present invention the
sample cells are subjected to the MS analysis at a specific time of
growth. The preferred sampling times are mid and/or end log phase
of the cell growth.
[0067] In a preferred embodiment of the present invention the raw
sample MS data obtained in step b) and/or the raw standard MS data
provided in step a) are signal processed by an operation selected
from the group consisting of baseline correction, normalisation,
alignment, filtering and cropping.
[0068] In a preferred embodiment, the first MS signal processing
method used in step c) is selected from the group consisting of
baseline correaction, normalisation, alignment, filtering and
cropping.
[0069] MS data profiles typically exhibit a varied baseline due to
issues such as chemical noise in the MALDI matrix and ion
overloading. This is undesirable when using data analysis
techniques to compare MS profiles as their utilised distance matrix
to measure the similarity between profiles. Therefore, in a
preferred embodiment of the present invention the raw sample and/or
standard MS data are signal processed by the operation of baseline
correction.
[0070] Another commonly observed phenomenon with MS profiles is the
variation in the amplitude of the ion intensities. This can be
caused by a number of factors, such as variation in the sample
preparation or changes in the sensitivity over the instrument.
Therefore, in a preferred embodiment of the present invention the
raw sample and/or standard MS data are signal processed by the
operation of normalisation.
[0071] Peak alignment is used to correct variation between the
observed M/Z value and the true time of flight. These errors
usually occur as a result of calibration errors and can be observed
as a systematic shift between peaks. Therefore, in a preferred
embodiment of the present invention the raw sample and/or standard
MS data are signal processed by the operation of peak
alignment.
[0072] Filtering of the MS profiles is carried out by smoothing the
signal preferably by a Savitzky-Golay filter. Therefore, in a
preferred embodiment of the present invention the raw sample and/or
standard MS data are signal processed by the operation of
filtering.
[0073] Cropping the MS profiles is performed to remove parts of the
signal containing little or no information. Preferably the range of
0 to 500 m/z units is removed from the MS spectra. Therefore, in a
preferred embodiment of the present invention the raw sample and/or
standard MS data are signal processed by the operation of
cropping.
[0074] In a preferred embodiment of the present invention the probe
of the at least one sample cell is re-sampled. This specific
preferred process step allows up-sampling and down-sampling of the
original signal, whilst preserving the information contained within
the spectra, i.e. altering the amounts channels of different data
points measured. Typically, re-sampling is utilised in situations
where the original high resolution MS signal would be considered
impractical to work with due to computational constraints such as
lack of computer memory. Re-sampling can also be used to create a
consistent m/z range, which facilitates lining up multiple
spectra.
[0075] In a preferred embodiment of the present invention the
sample of the at least one sample cell is up-sampled to at least
50,000, preferably to account for slight differences in m/z
vector.
[0076] In a preferred embodiment of the present invention the raw
standard and sample MS data are visually analysed.
[0077] In a preferred embodiment of the present invention the
pre-treated standard and sample MS profiles obtained in step c) are
optically analysed, preferably to detect outlier and to remove
unusual, defective pre-treated standard and sample MS profiles.
[0078] In a preferred embodiment, the first MS signal processing
method used in step c) comprises, preferably in the following
order, the following steps of resampling, baseline correction,
filtering, alignment, visual analysing and normalisation of the raw
standard and/or sample MS data.
[0079] In a preferred embodiment of the present invention the
sample cell with the predicted cell culture, preferably bioreactor,
performance data evaluated in step d) is cultivated in a cell
culture, preferably bioreactor, so as to verify its cell culture,
preferably bioreactor, performance data.
[0080] In a preferred embodiment of the present invention the raw
sample MS profiles obtained in step b) and the verified bioreactor
performance data of the sample cell are used in step a) as standard
MS data and bioreactor performance data from the standard cell. By
providing a higher number of MS profiles from standard cells with
their known cell culture, preferably bioreactor, performance data
the probability, preferably reliability, of the prediction is more
accurate and dependable.
[0081] In a preferred embodiment of the present invention the
PLS-DA model according to step (d) requires two different sets of
information, namely the x-block and the y-block. In a preferred
embodiment of the present invention the x-block contains the
information from within the pre-treated sample MS data generated at
the 96 DWP (deep well plate) stage of the process. In a preferred
embodiment of the present invention each pre-treated sample MS data
is treated as a sample, with the signal intensities recorded over a
specific range of m/z values being treated as the variables. In a
preferred embodiment of the present invention the y-block contains
information assigning each of the pre-treated standard MS data to a
class variable. In a preferred embodiment of the present invention
the y-block contains information preferably relating to specific
measures of productivity of a cell line at the bioreactor scale,
that means product concentration, specific productivity or integral
of viable cell count.
[0082] In a preferred embodiment of the present invention the
statistical program PLS-DA is not applied to differentiate stable
and unstable cell lines. In a preferred embodiment of the present
invention the statistical program PLS-DA is the only statistical
program used in the process according to the present invention. In
a preferred embodiment of the present invention the statistical
programs PLS-DA and PCA are the only statistical programs used in
the process according to the present invention. Preferably, the raw
standard and raw sample MS data are pre-treated in step c) without
the usage of a statistical analysis, preferably PLS, before being
subjected in step d) to a second MS signal processing method
including a PLS-DA-based comparative evaluation as to predict the
cell culture performance data of at least one sample cell.
[0083] In a preferred embodiment of the present invention using the
x-block data, a PLS mapping of the original variables into the
latent variable space is performed. This has the effect of reducing
the dimensionality of the problem, whilst describing as much of the
variability in the original data as possible. In a preferred
embodiment of the present invention the PLS-DA algorithm utilises
the information in the y-block to fit the linear discrimination
boundary that best separates the x-block data based on the class
information stored in the y-block. If there are only two classes
described in the y-block, a single discrimination boundary is
sufficient. In cases where three or more classes are present, the
within class samples should be compared to the out of class samples
for each available class.
[0084] In a preferred embodiment of the present invention the
following classes are used in the PLS-DA based comparative
evaluation to predict the integral viable cell count data: [0085]
High>4500.times.10.sup.6 cells.times.h/ml, [0086]
4500.times.10.sup.6
cells.times.h/ml>Medium>3250.times.10.sup.6 cells.times.h/ml
and [0087] Low<3250.times.10.sup.6 cells.times.h/ml.
[0088] In a preferred embodiment of the present invention the
following classes are used in the PLS-DA based comparative
evaluation to predict the cell specific productivity data: [0089]
High>2.35 pg.times.cell.times.h, [0090] 2.35
pg.times.cell.times.h>Medium>1.75 pg.times.cell.times.h and
[0091] Low<1.75 pg.times.cell.times.h.
[0092] In a preferred embodiment of the present invention the
following classes are used in the PLS-DA based comparative
evaluation to predict the cell product concentration data: [0093]
High>4 g/l and [0094] Low<4 g/l
[0095] The present invention relates also to a process for the
production of a cell product, preferably a protein, from a cell
comprising the steps: [0096] a) predicting the cell culture
performance data according to a method of the present invention,
[0097] b) identifying the sample cells with the desired cell
culture performance data, [0098] c) cultivating said sample cells
identified in step b), preferably in a large volume of a medium,
and [0099] d) obtaining the cell product, preferably the protein,
produced by the cultivation in step c).
[0100] Preferably, in the process for the production of a cell
product, preferably a protein, the sample cell which cell culture
performances data are predicted in step a) is cultivated in a low
volume of a medium.
[0101] Furthermore, the present invention provides a process for
the preparation, in particular isolation, of a cell with desired
cell culture, preferably bioreactor, performance data, wherein the
process for the prediction of cell culture, preferably bioreactor,
performance data of at least one sample cell is performed and the
at least one desired cell is prepared, preferably isolated.
[0102] The present invention provides a cell obtained, in
particular isolated, by a process according to the present
invention.
[0103] In a preferred embodiment of the present invention the cell
isolated is characterised by a protein productivity of at least 1
g/l, preferably at least 5 g/l, preferably at least 6, 7, 8, 9 or
preferably at least 10 g/l.
[0104] In a preferred embodiment of the present invention the cell
isolated is characterised by a protein productivity of at least 0.1
g/l/h, preferably at least 1 g/l/h, preferably at least 5 g/l/h,
preferably at least 6, 7, 8, 9 or preferably at least 10 g/l/h.
[0105] In a preferred embodiment of the present invention the cell
isolated is characterised by a protein productivity of at least 0.1
g/l/h/cell, preferably at least 1 g/l/h/cell, preferably at least 5
g/l/h/cell, preferably at least 6, 7, 8, 9 or preferably at least
10 g/l/h/cell.
[0106] The present invention solves its underlying problem also by
a device for the prediction of cell culture, preferably bioreactor,
performance data of at least one sample cell, preferably adapted to
providing a prediction of cell culture performance data, preferably
when supplied with a probe of at least one sample cell, comprising:
a) a means adapted for subjecting a probe of the at least sample
cell to a MS (mass spectrometric) analysis to obtain a raw sample
MS data thereof, (b) a means adapted for subjecting a raw standard
and the raw sample MS data to at least one first MS signal
processing method to obtain pre-treated standard and sample MS
profiles and (c) a means adapted for subjecting cell culture,
preferably bioreactor, performance data from a standard cell and
the pre-treated sample and standard MS profiles to a second MS
signal processing method including a PLS-DA (partial least square
discriminant analysis) based comparative evaluation so as to
predict the bioreactor performance data of the sample cell.
[0107] Further preferred embodiments are the subject matter of the
subclaims.
[0108] The present invention is further illustrated by way of the
following examples and the corresponding figures.
[0109] FIGS. 1a and 1b show a typical MS profile before and after
resampling to 1000 data points.
[0110] FIGS. 2a and 2b show typical MS profiles before and after
baseline correction.
[0111] FIGS. 3a and 3b show typical MS profiles before and after
normalisation.
[0112] FIGS. 4a and 4b show typical MS profiles before and after
Savitzky-Golay smoothing.
[0113] FIGS. 5a and 5b show typical MS profiles before and after
cropping regions of the spectra with little or no information.
[0114] FIGS. 6a, 6b and 38 show a score plot of PC1 versus PC2
using raw MS profiles (6a, 38) and with a typical associated MS
profile (6b).
[0115] FIGS. 7a, 7b and 39 show a score plot of PC1 versus PC2
using MS profiles subjected to baseline correction (7a, 39) with a
typical associated MS profile (7b).
[0116] FIGS. 8a, 8b and 40 show a score plot of PC1 versus PC2
using MS profile subjected to baseline correction and normalisation
(8a, 40) with a typical associated MS profile (8b).
[0117] FIGS. 9a, 9b and 41 show a score plot of PC1 versus PC2
using MS profiles subjected to baseline correction, normalisation
and cropping (9a, 41) with a typical associated MS profile
(9b).
[0118] FIG. 10 shows the PLSDA analysis flow chart in the
PLS_Toolbox.
[0119] FIGS. 11a and 11b show the PLSDA import dialogue box in the
PLS_Toolbox.
[0120] FIG. 12 shows PLSDA class group selection dialogue box in
the PLS_Toolbox.
[0121] FIG. 13 shows PLSDA data re-processing options in the
PLS_Toolbox.
[0122] FIG. 14 shows PLSDA cross validation dialogue box in the
PLS_Toolbox.
[0123] FIGS. 15a to 15d and 42 to 45 show various PLSDA scores
plots: (15a, 42) Bivariate scores plot, (15b, 43) Hotelling's T2
plot, (15c, 44) Model predictions plot and (15d, 45) Model
predictions probability plot.
[0124] FIGS. 16a, 16b, 46 and 47 show PLSDA loading plots: (16a,
46) latent variable 1 and (16b, 47) latent variable 2.
[0125] FIGS. 17 and 48 show PLSDA decision boundary plot for
Ypredicted Class 1.
[0126] FIGS. 18 and 49 show a PLSDA probability plot for Ypredicted
Class 1.
[0127] FIGS. 19a and 19b show PLSDA analysis comparing the
LC-ESI-MS data from the antibody-producing cell lines CHO 2, 42 and
52.
[0128] FIG. 20 shows PLSDA analysis comparing the LC-ESI-MS data
from seven antibody-producing CHO cell lines (2, 42, 52, 75, 106,
144 and 164) with samples grouped into lower than 2 g/l and higher
than 2 g/l.
[0129] FIGS. 21a and 21b show MS profiles pre-treated with a MS
signal processing method comprising baseline correction and
normalisation (21a) and comprising resampling, baseline correction,
filtering, alignment, visual analysing and normalisation (21b).
[0130] FIG. 22 shows a Y Predicted plot for cell lines Round 1
predictions using the pre-treatment of raw MS data according to the
present invention, including resampling, baseline correction,
filtering, alignment, visual analysing and normalisation of the raw
MS data, and subsequent PLS-DA modelling.
[0131] FIG. 23 shows a LV1 vs LV2 for cell lines Round 1
predictions using the pre-treatment according to the present
invention.
[0132] FIG. 24 shows a Y Predicted plot for cell lines Round 2
Predictions using the pre-treatment of raw MS data according to the
present invention, including resampling, baseline correction,
filtering, alignment, visual analysing and normalisation of the raw
MS data, and subsequent PLS-DA modelling.
[0133] FIG. 25 shows a LV1 vs LV2 for cell lines Round 2
predictions using the pre-treatment according to the present
invention.
[0134] FIG. 26 shows a Y Predicted plot for cell lines Round 3
Predictions using the pre-treatment of raw MS data according to the
present invention, including resampling, baseline correction,
filtering, alignment, visual analysing and normalisation of the raw
MS data, and subsequent PLS-DA modelling.
[0135] FIG. 27 shows a LV1 vs LV2 for cell lines Round 3
predictions using the pre-treatment according to the present
invention.
[0136] FIG. 28 shows a Y Predicted plot for cell lines Round 4
Predictions using the pre-treatment of raw MS data according to the
present invention, including resampling, baseline correction,
filtering, alignment, visual analysing and normalisation of the raw
MS data, and subsequent PLS-DA modelling.
[0137] FIG. 29 shows a LV1 vs LV2 for cell lines Round 4
predictions using the pre-treatment according to the present
invention.
[0138] FIG. 30 shows a Y Predicted plot for cell lines Round 5
Predictions using the pre-treatment of raw MS data according to the
present invention, including resampling, baseline correction,
filtering, alignment, visual analysing and normalisation of the raw
MS data, and subsequent PLS-DA modelling.
[0139] FIG. 31 shows a LV1 vs LV2 for cell lines Round 5
predictions using the pre-treatment according to the present
invention.
[0140] FIG. 32 shows an IVC Y Predicted plot for cell lines Round 3
Predictions using the pre-treatment of raw MS data according to the
present invention, including resampling, baseline correction,
filtering, alignment, visual analysing and normalisation of the raw
MS data, and subsequent PLS-DA modelling.
[0141] FIG. 33 shows an IVC Y Predicted plot for cell lines Round 4
Predictions using the pre-treatment of raw MS data according to the
present invention, including resampling, baseline correction,
filtering, alignment, visual analysing and normalisation of the raw
MS data, and subsequent PLS-DA modelling.
[0142] FIG. 34 shows an IVC Y Predicted plot for cell lines Round 5
Predictions using the pre-treatment of raw MS data according to the
present invention, including resampling, baseline correction,
filtering, alignment, visual analysing and normalisation of the raw
MS data, and subsequent PLS-DA modelling.
[0143] FIG. 35 shows a qP Y Predicted plot for cell lines Round 3
Predictions using the pre-treatment of raw MS data according to the
present invention, including resampling, baseline correction,
filtering, alignment, visual analysing and normalisation of the raw
MS data, and subsequent PLS-DA modelling.
[0144] FIG. 36 shows a qP Y Predicted plot for cell lines Round 4
Predictions using the pre-treatment of raw MS data according to the
present invention, including resampling, baseline correction,
filtering, alignment, visual analysing and normalisation of the raw
MS data, and subsequent PLS-DA modelling.
[0145] FIG. 37 shows a qP Y Predicted plot for cell lines Round 5
Predictions using the pre-treatment of raw MS data according to the
present invention, including resampling, baseline correction,
filtering, alignment, visual analysing and normalisation of the raw
MS data, and subsequent PLS-DA modelling.
EXAMPLE 1
Cell Line Generation (According to the State of the Art)
[0146] A GS expression vector (Lonza) containing gene-optimised
heavy and light chain genes for the expression of a model
mouse-human chimeric IgG4 or IgG1 antibody (Kalwy et al., Mol.
Biotechnol. 2006, 34, 151-156) was used to generate recombinant,
antibody expressing GS-CHO cell lines. The vector was introduced
into the host cell line, CHOK1SV (a derivative of CHO-K1; Lonza),
using standard electroporation methods and the transfection mixture
was distributed across eighty 96-well plates. Plates were incubated
at 37.degree. C. in a humidified, 10% CO.sub.2 in air atmosphere.
The following day, fresh medium was added to the cell suspension in
the plates. The MSX (methionine sulphoxamine) concentration in the
medium was such that the final MSX concentration in each well was
50 .mu.M. Plates were first screened for glutamine-independent
transfectants at approximately 3 weeks post transfection.
Transfectant colonies isolated (each identified as originating from
a well with a single colony) were progressed through all the
assessment stages of a typical cell line construction strategy.
[0147] Cell concentration of the cultures was determined using a
Vi-CELL.TM. automated cell viability analyser (Beckman Coulter).
Cultures were established in 125 mL shake-flasks with a target cell
concentration of 2.0.times.10.sup.5 viable cells/mL and a final
volume of typically 30 mL. Cell lines were serially subcultured on
a 4 day regime. Once acceptable cell concentrations at subculture
were reached and any large fluctuations in viable cell
concentration between subcultures had ceased, the assessment stages
performed in suspension culture commenced. The `fed-batch`
assessment was performed after the cell lines were ranked following
the first suspension evaluation (`batch`). For the fed-batch
assessment, the cell concentration of the cultures was determined
on days 7 and 14 using a Vi-CELL.TM. automated cell viability
analyser. A bolus addition of feed A was made on day 3 and bolus
additions of feed B were made on days 8 and 11. Samples of culture
supernatant were taken on different days for antibody concentration
determination. Cell viability analysis could alternatively be done
with MACSQuant.RTM. Analyzer.
EXAMPLE 2
Preparation of the Sample Cells for the MALDI-TOF Analysis
[0148] Unless otherwise specified, all experiments have been
conducted under the same culture conditions as outlined under the
example 1 (first paragraph). Before the sample probes (that means
the samples) were subjected to the MS analysis the cells were
counted and the required volume of culture to provide the
appropriate number of cells calculated. Cells were removed from the
incubator (96 well plate) immediately prior to processing. The
required volume of each sample was transferred to an Eppendorf
tube, centrifuged for 5 minutes at 960 rcf (3000 rpm) in an
Eppendorf microfuge (model 5417c, rotor F-45-30-11) and the
supernatant removed. The cells were then washed with 1 ml of PBS
(phosphate buffered saline) by gently pipetting up and down then
centrifuged as above. Where indicated, cells were subsequently
washed with 1 ml of 0.35 M sucrose and the supernatant removed
after centrifuging as described above. At that point cell pellets
could be stored (-80.degree. C.) for further handling in the future
or immediately processed for MS analysis. In case of storage frozen
cell pellets need to equilibrate to room temperature before used
after thawing.
[0149] A 20 mg/ml solution of sinapinic acid was prepared in matrix
buffer (40% acetonitrile, 60% 0.1% TFA) which results in a
saturated solution. The sinapinic acid solution was then placed in
a sonicating water bath for 15 minutes before centrifugation at
17900 rcf (13000 rpm) for 5 minutes in an Eppendorf microfuge
(model 5417c, rotor F45-30-11).
[0150] Matrix solution (50 .mu.l) was then added to each sample and
the cells re-suspended by manually pipetting the solution up and
down. After resuspension the cells were placed at 4.degree. C. for
up to several hours. On removal from 4.degree. C., the cells were
re-suspended by gently tapping the tube and then 1 .mu.l of each
sample was spotted onto a 384 MTP ground steel MALDI TOF plate
(Bruker). Samples were allowed to air dry before the plate was put
into the MALDI TOF machine (Bruker Ultraflex) and the samples
analysed.
EXAMPLE 3
Preparation of Sample Cells for LC-ESI-MS Analysis Using Dunn Lysis
Buffer
[0151] Sample collection: A range of CHO cell lines were grown in
250 ml suspension cell culture flasks. Cells were counted using a
Vi-CELL.TM. and the cell number required (1.times.10.sup.6 to
0.015625.times.10.sup.6) were pipetted into 1.5 ml Eppendorf tubes
and centrifuged at 960 rcf in an Eppendorf microfuge (model 5417c,
rotor F-45-30-11) for 5 mins and the supernatant removed. The
pellets were stored at -80.degree. C. until used.
[0152] Cell lysis: The pellets were thawed and resuspended in 400
.mu.l of Dunn Lysis buffer (Ultra pure urea 9.5 M, CHAPS 2%, DTT
1%) vortexed thoroughly and incubated at room temperature (RT) for
1 h with a brief vortex at 30 min after the start of incubation.
Samples were then centrifuged at 985.6 g, preferably 1700 rcf
(relative centrifugal force), for 1 min to remove cell debris and
the supernatant was pipetted into 2 ml Eppendorfs. 50 .mu.l of
sample was then used for acetone precipitation.
[0153] Acetone precipitation:A 4:1 dilution of 100% ice cold
acetone to sample was incubated for 1 h at -20.degree. C. The
diluted sample was then centrifuged at 8870.4 g, preferably 17900
rcf (relative centrifugal force), for 10 min, the supernatant
removed and the pellet left to dry at air briefly (not more than 5
min).
[0154] The 2D clean-up kit from GE healthcare (product code
80-6484-51) was used to clean up the samples before the solution
tryptic digest. Procedure A from the manual supplied with the kit
was followed.
[0155] Tryptic digest in solution: The pellet was re-suspended in
50 .mu.l of 8 M urea, 0.4 M ammonium bicarbonate
(NH.sub.4HCO.sub.3) by pipetting the sample up and down to
initially dislodge the pellet followed by brief vortexing. The
sample was reduced chemically by adding 2.5 .mu.l of 100 mM
dithiothreitol (DTT) in 50 mM NH.sub.4HCO.sub.3 for 1 h in a
37.degree. C. incubator. The sample was then alkylated by adding 5
.mu.l of 100 mM iodoacetamide in 50 mM NH.sub.4HCO.sub.3 for 15 min
at RT in the dark. The urea concentration was diluted to <2 M by
adding 192.5 .mu.l of HPLC grade water followed by the addition of
10 .mu.l of 0.25 .mu.g/ul modified trypsin (Promega). Tryptic
digestion was then left to proceed overnight in a 37.degree. C.
incubator. The sample was then dried down using a Savant speed vac
(SC110A) on a low setting and resuspended in 20 .mu.l of 0.1%
formic acid, centrifuged for 8870.4 g for 1 min, the supernatant
removed and any pellet resuspended and centrifuged again at 8870.4
g, preferably 17900 rcf (relative centrifugal force), for 1 min
then pipetted into screw cap vials with inserts and frozen at
-80.degree. C. until analysed by LC-ESI-MS.
EXAMPLE 4
Analysis with LC-ESI-MS
[0156] A HPLC method used for the analysis with LC-ESI-MS (ESI-MS
(Bruker or Waters) coupled with HPLC (Donex or Agilent)) is shown
in Table 1 that resulted in appropriate MS spectra.
[0157] The files produced by the LC-ESI-MS were then converted from
the proprietary file format (Bruker or Waters) to a universal
standard (mzXML) (i.e. using CompassExport) and the resulting files
and data subjected to a binning procedure. The binning approach,
which is standard for the analysis of this type of MS data, allows
the comparison of multiple ESI-MS datasets from different (or the
same) samples by aligning them and involves dividing the retention
time (elution time from LC system) and m/z range (mass to charge
ratio of ions as detected in ESI-MS) into equally spaced intervals,
for example, using a retention time bin of 60 seconds and a mass to
charge bin of 1 m/z unit per bin.
TABLE-US-00001 TABLE 1 Example HPLC gradient run. A 35 min gradient
with a flow rate of 0.3 .mu.l/min throughout the run using a
multistep gradient as displayed below with buffer A comprising 0.1%
formic acid and buffer B comprising 80% acetonitrile (ACN) and 0.1%
formic acid. Time % of buffer B Flow rate (.mu.l/min) 0 4 0.3 0 4
0.3 10 55 0.3 11 90 0.3 16 90 0.3 17 4 0.3 35 4 0.3
EXAMPLE 5
Data Analysis Protocol Method 2
5.1 Data Processing and Software Development
[0158] In the present example for MS based cell line screening and
generation, a software tool--run via a Windows interface--which
allows the fast and across scale prediction of cell line
productivity is used. It is compiled in MATLAB (release 2008b,
reference) using the MATLAB Bioinformatics and Statistics toolboxes
as well as the PLS_Toolbox (www.eigenvector.com).
[0159] The software application starts with the availability of MS
profiles from sample and standard cell lines having been grown
under different culture conditions and scales. The signal
processing tools have been applied to the MS profiles to extract
unique MS data patterns indicative of different levels of product
producing cell lines.
5.2 Re-Sampling MS Profiles
[0160] Re-sampling of MS profiles is performed using the
`msresample` function from the MATLAB Bioinformatics Toolbox
(http://tinyurl.com/msresample). This allows the up-sampling and
down-sampling of the original signal, whilst preserving the
information contained within the spectra. FIGS. 1a and 1b show a
typical 96DWP spectrum before and after application of
re-sampling.
[0161] Typically re-sampling is utilised in situations where the
original high resolution MS signal would be considered impractical
to work with due to computational constraints such as lack of
computer memory. Re-sampling can also be used to create a
consistent m/z range, which facilitates lining up multiple spectra.
Care must be taken when re-sampling MS profiles so as not to set
the number of re-sampled units too low. This will cause the signals
to lose resolution and can result in a loss of features.
5.3 Baseline Correction of MS Profiles
[0162] MS data profiles typically exhibit a varied baseline due to
issues such as chemical noise in the MALDI matrix and ion
overloading. This can be undesirable when using data analysis
techniques to compare MS profiles as they utilise distance metrics
to measure the similarity between profiles. It is therefore
preferred to remove these effects prior to any form of comparative
analysis of the signals. This is performed using the `msbackadj`
function in the MATLAB Bioinformatics Toolbox
(http://tinyurl.com/msbackadj). FIGS. 2a and 2b show a selection of
typical 96DWP spectra before and after application of baseline
correction.
[0163] When applying a number of spectral pre-treatments in series,
baseline correction should be used after down-sampling and prior to
correcting the calibration, as the noise present will impact on the
result.
5.4 Normalisation
[0164] Another commonly observed phenomenon with MS profiles is a
variation in the amplitude of the ion intensities. This can be
caused by a number of factors, such as variation in the sample
preparation or changes in the sensitivity of the instrument. The
standard procedure to account for this variation is to normalise
the area under the MS curves to that of the group average
(typically the mean or median is used). This is performed using the
`msnorm` function from the MATLAB Bioinformatics Toolbox
(http://tinyurl.com/msnormal). FIGS. 3a and 3b show a selection of
typical 96DWP spectra before and after normalisation of the area
under the curve.
[0165] When applying a number of spectral pre-treatments in series,
normalisation of the samples should be performed after subtracting
the baseline as the noise element introduced by the crystallisation
matrix can impact on the results.
5.5 MS Alignment
[0166] Peak alignment is used to correct variation between the
observed m/z value and true time of flight. These errors usually
occur as a result of calibration errors and can be observed as a
systematic shift between peaks. Correction of these inconsistencies
can be performed using the `msalign` function from the MATLAB
Bioinformatics Toolbox (http://tinyurl.com/msalign).
[0167] One method to align spectra is to spike the samples with a
substance with a known spectral profile, and align the samples
based on this. However, in situations where the samples have not
been spiked, samples can be aligned relative to reference spectra
such as the mean profile.
5.6 Filtering of MS Profiles
[0168] A typical MS profile contains a mixture of both signal and
noise. Smoothing of the signal by use of a Savitzky-Golay filter
can help to reduce the impact of the noise component of the signal
during subsequent processing. Savitzky-Golay filters are typically
applied to MS signals as they use high order polynomials to fit the
curves. This results in greater preservation of the features in the
signal, such as the peak heights. This process is performed using
the `mssgolay` function from the MATLAB Bioinformatics Toolbox
(http://tinyurl.com/mssgolay). FIGS. 4a and 4b show a selection of
typical 96DWP spectra before and after application of
Savitzky-Golay filtering.
5.7 Cropping of MS Profiles
[0169] Cropping of the MS profiles is performed to remove parts of
the signal containing little or no information. It also allows the
spectra to be divided into subsections. This enables specific
regions of the MS profiles to be analysed rather than the whole
spectra. FIGS. 5a and 5b show a selection of typical 96DWP spectra
before and after cropping of the signal in the range of from 0 to
500 m/z.
5.8 Comparing the Effects of Pre-Treatment
[0170] To demonstrate the effect of applying signal processing
techniques to MS data, a group of 118 cell lines (measured in
duplicate) were analysed using Principal Component Analysis (PCA).
FIGS. 6a, 6b and 38 show the principal component scores plot of PC1
vs. PC2. This plot describes the two major sources of variation in
the data set.
[0171] Application of baseline correction to the raw MS profiles
results in a reduction of the amount of scatter observed in the
first and second principal components. This can be observed in
FIGS. 7a, 7b and 39.
[0172] Baseline correction only accounts for the noise in the
signal due to the MALDI matrix. It is preferred that the variation
in the amplitudes of the signal be removed using normalisation.
FIGS. 8a, 8b and 40 show the effect of applying normalisation on
the group of spectra. It clearly shows a reduction in the variation
observed in PC1, the major source of variation.
[0173] The final signal processing step performed is to remove
parts of the signal known to contain no useful information. FIGS.
9a, 9b and 41 show the effect of removing the data points in the MS
profile over the range from 0 to 500 m/z units, as the MALDI was
set not to record intensities in this range. It is clear from the
scores plot that no effect is observed by removing this data, as
the scores plot is identical to that observed in FIGS. 8a 8b and
40.
[0174] Using in step c) a first signal processing method comprising
the following steps of resampling, baseline correction, filtering,
alignment, visual analysing and normalisation of the raw standard
and/or sample MS data, the standard and sample MS profiles appear
smoother and the peaks align more consistently across cell lines
compared to MS profiles pretreated solely with the method steps of
baseline correaction and normalisation (FIGS. 21a and 21b). Said
method also results in more consistent MS profiles for biological
replicates where 2 or 3 cell pellets were prepared for subjecting
to MS analysis from the same sample. Preferably, these improvements
do come at the cost of an increase in the time required to process
the samples (approx. 2 h for approx. 400 spectra), however this is
not preferably prohibitive in terms of a modelling approach and
time required to predict/select those cell lines of interest during
a cell line construction process.
EXAMPLE 6
Signal Processing Method 2
6.1 PLSDA Modelling of Productivity Metrics
[0175] PLSDA is an application of multivariate least squares
modelling specifically formulated for predictive classification.
The developed MS fingerprinting approach utilised in the example
employs the PLS_Toolbox implementation of PLSDA, published by
Eigenvector Research, Inc. (EVRI) (www.eigenvector.com).
6.2 Training a PLSDA Model
[0176] To train a PLSDA model, two different sets of information
are required; the x-block and the y-block. In the outlined approach
to performing a new cell line construction, the x-block contains
the information from within the spectral profiles generated at the
96DWP stage of the process. Each profile is treated as a sample,
with the signal intensities recorded over a specific range of m/z
values being treated as the variables. The y-block contains
information assigning each of the training samples to a class
variable. In this example the y-block contains information relating
to specific cell culture data of productivity of a cell line at the
bioreactor scale, i.e. product concentration, specific productivity
or integral of viable cell count.
[0177] Using the x-block data, a PLS mapping of the original
variables into the latent variable space is performed. This has the
effect of reducing the dimensionality of the problem, whilst
describing as much of the variability in the original data as
possible. The PLSDA algorithm then utilises the information in the
y-block to fit the linear discrimination boundary that best
separates the x-block data based on the class information stored in
the y-block. If there are only two classes described in the
y-block, a single discrimination boundary is sufficient; in cases
where three or more classes are present, the within class samples
should be compared to the out of class samples for each available
class.
6.3 Analysis Flowchart
[0178] FIG. 10 shows the flowchart of operations required to build
a PLSDA model using the graphical implementation of the algorithm
found in the MATLAB PLS_Toolbox.
1. Load X data--This button prompts the user to import the x-block
data into the software. The fingerprinting software, ms_preproc, is
designed to act as an interface between the signal processing and
analysis techniques, automatically converts the MS data into the
required format to work with the PLS_Toolbox and saves the variable
to the MATLAB workspace as the variable `Xblock` (FIGS. 11a and
11b). 2. Load Classes (optional)--This button is used to import the
y-block data into the software. However, class information can also
be stored in the `Xblock` variable (see
http://wiki.eigenvector.com/index.php?title=DataSet_Object for more
details). The ms_preproc software embeds the class information into
the `Xblock` variable; hence this step is optional. 3. Select Class
Groups--This button presents the user with the option to select the
class groups with which to build the model. FIG. 12 shows a typical
example with three classes (high, medium, low). The model will
calculate decision boundaries for the classes added to the
right-hand column. 4. Choose Preprocessing--This button can be used
to apply various preprocessing techniques to the spectra files
(FIG. 13). If this step is performed using the spectral
pre-processing tools in the bioinformatics toolbox, the majority of
this step is optional. However, also related to preprocessing of
the data is the issue of the prior probabilities assigned during
training. By default the algorithm assume an equal probability that
each class is selected. Sometime this is not the case, and the
values must be adjusted to reflect the true probabilities. This is
performed by altering the `Method Options (PLSDA)`, which is under
the `Options` section of the `Edit` menu. 5. Choose
Cross-Validation--This button allows the user to cross validate
during the training process. This process is often used to provide
an improved degree of confidence in a result and serves as a
cross-check of classifier performance. The standard approach is to
reserve a portion of the training data and use this to test the
performance of the classifier. Typically, this process is then
repeated with a different portion of data reserved. FIG. 14 shows
the methods available for portioning the data using the PLS_toolbox
(http://wiki.eigenvector.com/index.php?title=Using_Cross-Validation).
6. Build Model--This button calculates the PLS latent variables and
places the optimum position of the discrimination boundaries so as
to maximise the number of correct classifications within the
training dataset. 7. Choose Components--This button becomes
available once the PLSDA model has been calculated. It produces
graphs to aid the user in the selection of the number of Latent
Variables to retain in the PLSDA model. Another method to achieve
this is to assume that the variation explained relative to the
y-block should be in the region of 70-80% so as not to over-fit the
model. 8. Review Scores--This button allows access to a number of
plots related to the latent variable scores. These can be used to
analyse the model performance. FIGS. 15a to 15d and 42 to 45 show
some of the most useful graphs; (a) Bivariate Loadings Plot, (b)
Hotelling's T2 plot, (c) Model Predictions Plot and (d) Associated
Probabilities for the Model Predictions Plot. 9. Review
Loadings--This button allows access to a number of plots related to
the latent variable loadings. These can be used to identify the
variables that have the most significant influence on each latent
variable. This can be useful for identifying the areas of the
spectral signal that are the most likely to influence
discrimination. 10. Load Test Data--Once the model has been built,
the next step is to utilise the models to make predictions about
unseen data. Clicking this button presents the user with the same
dialogue box as in FIGS. 11a/b. Test data can also be loaded in
with the original x-block data in step 1 by not assigning a class
variable to the samples. Any samples not assigned a class by our
ms_preproc software during step 2 are automatically regarded as
test data. These can be seen as the samples classed as unknown in
FIGS. 16a, 16b, 46 and 47. 11. Apply model--This button fits the
test data to the trained PLSDA model. It allows the user to
determine the most probable class to which unknown samples will
belong.
EXAMPLE 7
MS Profiles Subjected to MALDI-TOF and their Statistical
Modelling
[0179] The paragraph exemplified focuses on results obtained from
the modelling of the MS analysis data during the new cell line
generation process. FIGS. 48, 49, 17 and 18 show the results
obtained for the BNCD model (data are baseline corrected,
normalised and cropped with duplicate samples included). The
triangles (FIGS. 17 and 18) and stars (FIGS. 48 and 49) show the
training data for the "High" class (>4000 mg/L), stars (FIGS. 17
and 18) and triangles (FIGS. 48 and 49) show the training samples
for the "Low" class (.ltoreq.3999 mg/L), black dots (FIGS. 17 and
18) and crosses (FIGS. 48 and 49) without a cell line ID number
represent samples for which the class data is unknown and dots with
cell line ID number show the samples that fall above the
classification boundary (upper grey dotted line).
[0180] Using the processed information from the cell line
construction process (i) a prediction model could be built
including hundreds of MS data generated during the cell line
generation process. Based on the model a list of the cell lines
that were expected to produce different amount of MAb (>4000
mg/L; .ltoreq.3999 mg/L) was collated. Table 2 highlights several
cell lines which can be identified in FIGS. 48, 49, 17 and 18. The
cell lines were grown with their titre values recorded to measure
the performance of the prediction method (Table 2).
TABLE-US-00002 TABLE 2 Predicted high/low producing cell lines vs.
observed productivity Cell line ID Prediction 10 L bioreactor 262B7
>4000 mg/L 6524 mg/L 281D8 >4000 mg/L 4555 mg/L 241B6
.ltoreq.3999 mg/L 1219 mg/L 243D11 .ltoreq.3999 mg/L 663 mg/L 246F9
.ltoreq.3999 mg/L 964 mg/L
[0181] The results of the validation run proofed the successful
application of predictive cell line selection in the process of
cell line generation, so that at early development stage collated
MS profiles of individual cell lines reflect their behaviour at the
later manufacturing scale.
EXAMPLE 8
The Results after the Statistical Modelling of MS Profiles
Subjected to LC-ESI-MS
[0182] The raw MS profiles obtained by LC-ESI-MS were signal
processed by signal processing method I (example 5) and signal
processing method II (example 6). The results were shown in FIGS.
19a/b and 20.
[0183] FIGS. 19a/b show the separation of three different
recombinant CHO cell lines using a PLSDA analysis (FIGS. 19a/b show
LV1 v LV3) and FIG. 20 shows the separation of seven different
recombinant CHO cell lines after PLSDA analysis when the samples
are grouped into <2 g/L and >2 g/L. The data shows two
samples of each recombinant cell line group and that the PLSDA
algorithm is capable of discriminating between cell lines belonging
to different groups. The approach is suitable for fingerprinting
recombinant cell lines on the basis of desirable (e.g.
productivity) characteristics.
[0184] Table 3 shows the product concentration of the CHO cell
lines 2, 42, 52, 75, 106, 144 and 164 cultivated in a 24 well
plate, batch, fed batch and bioreactor. Especially, the product
concentration at bioreactor scale was predicted correctly by the
PLS-DA analysis using LC-ESI-MS data (FIG. 20).
TABLE-US-00003 TABLE 3 Product Grouping as Product Product Product
conc. shown in PLS- CHO conc. 24 conc. conc. Fed bioreactor DA
analysis Cell well plate Batch batch scale of LC-ESI-MS line (mg/L)
(mg/L) (mg/L) (mg/L) data 42 230 538 2404 3220.00 >2 g/L 52 31.5
31.5 101 24.00 <2 g/L 2 236 480 1680.5 2594.00 >2 g/L 144 175
391 1592 1816.00 <2 g/L 75 241 606 1001 1826.00 <2 g/L 106
221 766 969.5 2325.00 >2 g/L 164 202 534 881.5 2307.00 >2
g/L
EXAMPLE 9
Improving the Accuracy of Prediction
9.1 Bioreactor Round 1
[0185] In this first round, cell lines producing the antibody IgG
XXX anti insulin have been conducted, unless otherwise specified,
under the culture conditions as outlined in Example 1, first
paragraph.
[0186] Then, said cell lines were subjected to a MS analysis to
obtain raw sample MS data thereof, subsequently, the raw sample MS
data were subjected to at least one first MS signal processing
method comprising the steps of up-sampling, baseline correction,
filtering, alignment and normalization to obtain pre-treated sample
MS profiles and then the pre-treated sample MS profiles were
subjected to a second data analysis comprising of a PLS-DA based
comparative evaluation so as to predict the titre data in a
bioreactor of the cell lines.
[0187] As pre-treated standard MS profiles MS data of cell lines
producing the antibody IgG CB72.3 and pre-treated by baseline
correction, normalisation and cropping, especially pretreated MS
profiles of the cell lines in Table 2 (Example 7), are used. As
titre data from a standard cell the titre data from cell lines
producing the antibody IgG CB72.3, especially the titre data listed
in Table 2 (Example 7), are used.
[0188] The Y predicted plot obtained by said method shows a set of
cell lines would have been considered for bioreactor evaluation
(FIG. 22). There is also a significant tail off of the validation
data on the latent variables plot (FIG. 23) indicating that there
may not be sufficient coverage of the data space in the model to
accurately predict the titre of all cell lines.
[0189] Afterwards, the cell lines with the predicted titre data are
cultivated in a cell culture so as to verify its titre data in a
bioreactor. The bioreactor cultivation is carried out in a
conventional manner.
[0190] The samples of the different cell lines were taken on the
15th day of bioreactor cultivation.
[0191] Table 4 shows the resultant titre data of the cell lines
that were actually run in the first round of cultivation in a
bioreactor.
TABLE-US-00004 TABLE 4 Bioreactor Prediction from the Cell titre
data method according to Line (mg/L) the present invention 025G12
7224 1/3 reps > 4 mg/ml 929H9 7385 1/3 reps > 4 mg/ml 897G3
2452 0/3 reps > 4 mg/ml The term "reps" means the repetitions of
the preparations of one cell line to be subjected to MS
analysis.
[0192] Under the term "x/3 reps>4 mg/ml", wherein x can be 0, 1,
2 or 3, is understood that in x cases of the three preparations of
one cell line the PLS-DA based comparative evaluation predicts a
titre data of more than 4 mg/ml.
9.2 Bioreactor Round 2
[0193] Bioreactor Round 2 has been carried out in the same way as
specified in Example 9.1. However, as standard pre-treated MS
profiles the pre-treated standard MS profiles of the Bioreactor
Round 1 and the MS profiles of the cell lines, which titre data has
been measured in the first run of bioreactors, have been included
together with its measured titre data in the statistical
program.
[0194] The Y predicted plot obtained by said method shows a
different set of cell lines that would have been considered for
bioreactor cultivation (FIG. 24) compared to FIG. 22. The tail off
in the validation data is less pronounced on the latent variables
plot compared to FIG. 23 (FIG. 25). However, it is still pronounced
and the method implies a large number of cell lines will be high
producers.
[0195] Table 5 shows the resultant titres of the cell lines that
were actually run in the second round of bioreactors.
TABLE-US-00005 TABLE 5 Bioreactor Prediction from the Cell titre
data method according to Line (mg/L) the present invention 906G5
1311 1/2 reps > 4 mg/ml 930C4 108 3/3 reps > 4 mg/ml 934H6
1232 2/2 reps > 4 mg/ml 952D9 79.1 2/2 reps > 4 mg/ml 920D6
1911 3/3 reps > 4 mg/ml 964E7 7591 2/3 reps > 4 mg/ml
9.3 Bioreactor Round 3
[0196] Bioreactor Round 3 has been carried out in the same way as
specified in Example 9.1. However, as standard pre-treated MS
profiles the pre-treated standard MS profiles of the Bioreactor
Round 1 and the MS profiles of the cell lines, which titre data has
been measured in the first and second run of bioreactors, together
with their measured titre data, have been included in the
statistical program.
[0197] The Y predicted plot obtained by said method predicts that
less cell lines will be high producers (FIG. 26). The tail off in
the validation data shown in the latent variables plot (FIG. 27) is
significantly reduced compared to FIG. 25. The prediction accuracy
appears more reliable than the previous 2 rounds implying that the
method is better fit to the data space than the previous ones.
[0198] Table 6 shows the resultant titres of the cell lines that
were actually run in the third round of bioreactors.
TABLE-US-00006 TABLE 6 Bioreactor Prediction from the Cell titre
data method according to Line (mg/L) the present invention 029D11
3074 1/3 reps > 4 mg/ml 906B8 478 0/3 reps > 4 mg/ml 917C3
2451 1/2 reps > 4 mg/ml 946C4 823 0/3 reps > 4 mg/ml 961H8
5660 2/2 reps > 4 mg/ml 952C8 3959 2/3 reps > 4 mg/ml
9.4 Bioreactor Round 4
[0199] Bioreactor Round 4 has been carried out in the same way as
specifled in Example 9.1. However, as standard pre-treated MS
profiles the pre-treated standard MS profiles of the Bioreactor
Round 1 and the MS profiles of the cell lines, which titre data has
been measured in the first, second and third run of bioreactors,
together with their measured titre data, have been included in the
statistical program.
[0200] The Y predicted plot obtained by said method predicts fewer
of the cell lines will be high producers (FIG. 28) compared to FIG.
26. The tail off in the validation data on the latent variables
plot (FIG. 29) is again smaller that previously observed (for
instance FIG. 27).
[0201] Table 7 shows the resultant titres of the cell lines that
were actually run in the fourth round of bioreactors.
TABLE-US-00007 TABLE 7 Bioreactor Prediction from the Cell titre
data method according to the Line (mg/L) present invention 896C7
5132 2/2 reps > 4 mg/ml 931F12 1448 2/2 reps > 4 mg/ml 933A8
731 0/2 reps > 4 mg/ml 980F3 2428 0/2 reps > 4 mg/ml 917G3
4463 0/2 reps > 4 mg/ml 952F10 2083 2/2 reps > 4 mg/ml
9.5 Bioreactor Round 5
[0202] Bioreactor Round 5 has been carried out in the same way as
specified in Example 9.1. However, as standard pre-treated MS
profiles the pre-treated standard MS profiles of the Bioreactor
Round 1 and the MS profiles of the cell lines, which titre data has
been measured in the first, second, third and fourth run of
bioreactors, together with their measured titre data, have been
included in the statistical program.
[0203] The Y predicted plot obtained by said method shows only a
few cell lines being predicted to be high producers (FIG. 30). The
tail off in the validation data on the latent variables plot (FIG.
31) did not show much variation from that of FIG. 29. Of the 5 cell
lines that were run in this round, the model correctly classified 5
from 5, with 2 of the cell lines predicted to be high
producers.
[0204] Table 8 shows the resultant titres of the cell lines that
were actually run in the fifth round of bioreactors.
TABLE-US-00008 TABLE 8 Bioreactor Prediction from the Cell titre
data method according to Line (mg/L) the present invention 033D5
6024 2/2 reps > 4 mg/ml 016F11 2519 0/2 reps > 4 mg/ml 016B5
116 0/2 reps > 4 mg/ml 033G5 4155 2/3 reps > 4 mg/ml 948G2
1592 0/2 reps > 4 mg/ml
EXAMPLE 10
[0205] The cell lines were prepared and cultivated as outlined in
Example 9.
[0206] The term "Bioreactor Round X", where X is 3, 4 and 5, means
the same run of a bioreactor under the same cultivation conditions
for both examples 9 and 10.
[0207] For the PLS-DA based comparative evaluation a 3 class PLS-DA
model was built using both the integral viable cell count (IVC) and
the cell specific productivity (qP) data. In both models the
desired class was the Medium class. The class boundaries were
defined as follows:
IVC Model:
[0208] High>4500.times.10.sup.6 cells.times.h/ml
4500.times.10.sup.6
cells.times.h/ml>Medium>3250.times.10.sup.6
cells.times.h/ml
Low<3250.times.10.sup.6 cells.times.h/ml
qP Model:
[0209] High>2.35 pg.times.cell.times.h
2.35 pg.times.cell.times.h>Medium>1.75
pg.times.cell.times.h
Low<1.75 pg.times.cell.times.h
[0210] Based on the conditions and settings mentioned above a
PLS-DA based comparative evaluation was performed to predict the
cell culture performance data, namely the integral viable cell
count data (IVC) and the cell specific productivity data (qP).
[0211] Additionally, the IVC and qP data of the cell lines were
determined, when the cell lines were cultivated in the
bioreactor.
10.1 Prediction of the Integral Viable Cell Count Data (IVC)
[0212] As pre-treated standard MS profiles for the prediction of
Bioreactor Rounds 3 to 5 MS data of cell lines producing the
antibody IgG CB72.3 and pre-treated by baseline correction,
normalisation and cropping, especially pre-treated MS profiles the
cell lines in Table 2 (Example 7), and pre-treated MS profiles of
cell lines cultivated in the Bioreactor Rounds 1 and 2 are used. As
integral viable cell count data from a standard cell the integral
viable cell count data from cell lines producing the antibody IgG
CB72.3, especially from the cell lines in Table 2 (Example 7), and
the integral viable cell count data of the cell lines cultivated in
the Bioreactor Rounds 1 and 2 are used.
10.1.1 Bioreactor Round 3
[0213] FIG. 32 shows an IVC Y Predicted plot for cell lines Round 3
Predictions using the pre-treatment of raw MS data according to the
present invention and subsequent PLS-DA modelling.
[0214] Table 9 shows the observed IVC values of the cell lines that
were actually run in the third round of bioreactors.
TABLE-US-00009 TABLE 9 Prediction from the IVC data method
according to Cell (.times.10.sup.6 cells .times. the present
invention Line h/ml) (.times.10.sup.6 cells .times. h/ml) 029D11
1417 4500 > 1/3 reps > 3250 906B8 1686 4500 > 3/3 reps
> 3250 917C3 2748 4500 > 2/3 reps > 3250 946C4 3107 4500
> 1/3 reps > 3250 961H8 3873 4500 > 3/3 reps > 3250
952C8 2039 4500 > 0/2 reps > 3250
[0215] Under the term "4500>x/3 reps>3250", wherein x can be
0, 1, 2 or 3, is understood that in x cases of the three
preparations of one cell line the PLS-DA based comparative
evaluation predicts a IVC data between 4500 and 3250.
10.1.2 Bioreactor Round 4
[0216] FIG. 33 shows an IVC Y Predicted plot for cell lines Round 4
Predictions using the pre-treatment of raw MS data according to the
present invention and subsequent PLS-DA modelling.
[0217] Table 10 shows the observed IVC values of the cell lines
that were actually run in the fourth round of bioreactors.
TABLE-US-00010 TABLE 10 Prediction from the IVC data method
according to the Cell (.times.10.sup.6 cells .times. present
invention Line h/ml) (.times.10.sup.6 cells .times. h/ml) 896C7
4043 4500 > 2/2 reps > 3250 931F12 2730 4500 > 1/2 reps
> 3250 933A8 1812 4500 > 0/2 reps > 3250 980F3 3067 4500
> 1/2 reps > 3250 917G3 4057 4500 > 0/2 reps > 3250
952F10 3026 4500 > 1/2 reps > 3250
10.1.3 Bioreactor Round 5
[0218] FIG. 34 shows an IVC Y Predicted plot for cell lines Round 5
Predictions using the pre-treatment of raw MS data according to the
present invention and subsequent PLS-DA modelling.
[0219] Table 11 shows the observed IVC values of the cell lines
that were actually run in the fifth round of bioreactors.
TABLE-US-00011 TABLE 11 Prediction from the IVC data method
according to Cell (.times.10.sup.6 cells .times. the present
invention Line h/ml) (.times.10.sup.6 cells .times. h/ml) 033D5
4859 4500 > 2/2 reps > 3250 016F11 3472 4500 > 1/2 reps
> 3250 016B5 3553 4500 > 0/2 reps > 3250 033G5 3125 4500
> 3/3 reps > 3250 948G2 2008 4500 > 0/2 reps > 3250
10.2 Prediction of the Cell Specific Productivity Data (qP)
[0220] As pre-treated standard MS profiles for the prediction of
Bioreactor Rounds 3 to 5 MS data of cell lines producing the
antibody IgG CB72.3 and pre-treated by baseline correction,
normalisation and cropping, especially the pre-treated MS profiles
of the cell lines in Table 2 (Example 7), and pre-treated MS
profiles of cell lines cultivated in the Bioreactor Rounds 1 and 2
are used. As cell specific productivity data from a standard cell
the cell specific productivity data from cell lines producing the
antibody IgG CB72.3, especially from the cell lines in Table 2
(Example 7), and the cell specific productivity data of the cell
lines cultivated in the Bioreactor Rounds 1 and 2 are used.
10.2.1 Bioreactor Round 3
[0221] FIG. 35 shows a qP Y Predicted plot for cell lines Round 3
Predictions using the pre-treatment of raw MS data according to the
present invention and subsequent PLS-DA modelling.
[0222] Table 12 shows the observed qP values of the cell lines that
were actually run in the third round of bioreactors.
TABLE-US-00012 Prediction from the qP data method according to Cell
(pg .times. cell .times. the present invention Line h) (pg .times.
cell .times. h) 029D11 2.17 2.35 > 3/3 reps > 1.75 906B8 0.28
2.35 > 1/3 reps > 1.75 917C3 0.89 2.35 > 1/3 reps >
1.75 946C4 0.26 2.35 > 0/3 reps > 1.75 961H8 1.46 2.35 >
3/3 reps > 1.75 952C8 1.94 2.35 > 1/3 reps > 1.75
[0223] Under the term "2.35>x/3 reps>1.75", wherein x can be
0, 1, 2 or 3, is understood that in x cases of the three
preparations of one cell line the PLS-DA based comparative
evaluation predicts a qP data between 2.35 and 1.75.
10.2.2 Bioreactor Round 4
[0224] FIG. 36 shows a qP Y Predicted plot for cell lines Round 4
Predictions using the pre-treatment of raw MS data according to the
present invention and subsequent PLS-DA modelling.
[0225] Table 13 shows the observed qP values of the cell lines that
were actually run in the fourth round of bioreactors.
TABLE-US-00013 TABLE 13 Prediction from the qP data method
according to the Cell (pg .times. cell .times. present invention
Line h) (pg .times. cell .times. h) 896C7 1.27 2.35 > 2/2 reps
> 1.75 931F12 0.53 2.35 > 1/2 reps > 1.75 933A8 0.40 2.35
> 2/2 reps > 1.75 980F3 0.79 2.35 > 0/2 reps > 1.75
917G3 1.10 2.35 > 0/2 reps > 1.75 952F10 0.69 2.35 > 2/2
reps > 1.75
10.2.3 Bioreactor Round 5
[0226] FIG. 37 shows a qP Y Predicted plot for cell lines Round 5
Predictions using the pre-treatment of raw MS data according to the
present invention and subsequent PLS-DA modelling.
[0227] Table 14 shows the observed qP values of the cell lines that
were actually run in the fifth round of bioreactors.
TABLE-US-00014 TABLE 14 Prediction from the qP data method
according to Cell (pg .times. cell .times. the present invention
Line h) (pg .times. cell .times. h) 033D5 1.24 2.35 > 0/2 reps
> 1.75 016F11 0.73 2.35 > 1/2 reps > 1.75 016B5 0.03 2.35
> 0/2 reps > 1.75 033G5 1.33 2.35 > 2/3 reps > 1.75
948G2 0.79 2.35 > 0/2 reps > 1.75
10.3 Summary
[0228] The FIGS. 32 to 37 and the Tables 9 to 14 show that the IVC
and qP bioreactor performance data of cell lines cultivated in a 96
deep well plate can be predicted in an accurate way by the method
according to the present invention.
* * * * *
References